New AI Model Suggests Fats and Proteins Matter More for Blood Sugar Prediction Than We Thought

For decades, people with Type 1 diabetes have been taught to count carbohydrates to estimate insulin doses and predict blood sugar changes after meals. Carbohydrates do have the most immediate effect on blood glucose, which is why they're the standard focus. But a growing body of research suggests the picture is incomplete.

Fats and proteins also influence blood sugar levels—they may slow digestion, delay glucose absorption, and affect how long blood sugar stays elevated. However, these nutrients are rarely factored into glucose forecasting tools or meal-planning strategies because they're harder to account for consistently.

Scientists at multiple institutions developed a new artificial intelligence model called a temporal fusion transformer (TFT) to test whether including detailed nutritional information could improve blood sugar predictions. They trained the model on continuous glucose monitoring data, insulin doses, and meal information from 351 adults with Type 1 diabetes—representing over 15,600 individual meals.

The model was designed to predict blood glucose levels up to 4 hours after eating. Researchers tested different combinations: carbohydrates alone (the current standard), and then carbohydrates plus fats, proteins, sugars, and complex carbohydrates. This let them see whether additional nutritional data actually made predictions more accurate.

One of the key findings was that the importance of different nutrients changed depending on timing. In the first 2 hours after eating—when blood sugar rises most sharply—some nutrients were more predictive than others. In the 2–4 hour window, the contribution of each nutrient shifted again.

This suggests that blood sugar doesn't respond to food as a simple equation. The actual timing and interaction of carbohydrates, fats, and proteins create a more complex pattern that varies from person to person and meal to meal. The AI model was able to detect and learn these patterns from the data.

This is fundamental research aimed at understanding how nutrients influence blood glucose. The model's findings don't immediately translate into new meal-planning advice or changes to insulin dosing. The researchers were mapping out whether better data leads to better predictions—and they found evidence that it does.

The next steps would involve testing whether insights from this model could be simplified into practical tools that help people with Type 1 diabetes make real-time decisions about meals and insulin. For now, people with Type 1 diabetes should continue following their care team's guidance on carbohydrate counting and meal management. As with all emerging research, further validation and real-world testing would be needed before any changes to standard practice.

Source: JMIR AI. Evidence type: PubMed indexed literature. Type1Cure is an information and intelligence hub, not a medical advice service. This article summarizes published research and does not provide diagnosis, treatment, or personal medical guidance. Always talk to your own care team before changing anything about your Type 1 diabetes management.